The Agent Skills Directory

[PERSISTENCE_MECHANISMS]: The skill is designed to establish long-term, unattended operation by configuring and installing system-level services. It provides detailed instructions for generating and loading macOS LaunchAgents (.plist files) using launchctl bootstrap and managing systemd units on Linux.
[PRIVILEGE_ESCALATION]: The skill requires modifying system-level service states and interacting with service managers (launchctl, systemctl), which often involve administrative privileges.
[COMMAND_EXECUTION]: The skill executes multiple system commands to manage and verify services, including plutil -lint, launchctl print, and various CLI calls to the hermes tool.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill defines an "Alert relay contract" that exfiltrates system and task status data to external platforms such as Telegram using the hermes send CLI tool.
[DYNAMIC_EXECUTION]: The skill generates service configuration files at runtime, which are then loaded and executed by the operating system's service manager.
[INDIRECT_PROMPT_INJECTION]: The skill processes task instructions and context from an external SQLite database (Kanban board), which presents a surface for indirect prompt injection.
Ingestion points: Data is read from SQLite databases via board-scoped CLI commands such as boards list and show --json (SKILL.md).
Boundary markers: The instructions do not specify any delimiters or safety markers to differentiate between the agent's instructions and the data retrieved from the database.
Capability inventory: The skill possesses capabilities for file system writes (generating plists), shell command execution (launchctl, systemctl), and network operations (hermes send).
Sanitization: There is no mention of sanitizing or escaping the natural language content (task bodies) retrieved from the database before it is processed by the agent.

hermes-overnight-autonomy